Anthropic launches preview of Claude Duet proactive AI agent
The proactive assistant sends daily briefs, monitors connected apps, and performs tasks while seeking user approval for critical actions.

Anthropic enters proactive AI agent market with Claude Duet
Anthropic has unveiled a research preview of Claude Duet, an artificial intelligence agent designed to carry out work autonomously for professionals and small business owners.
Unlike traditional AI assistants that wait for explicit prompts, Claude Duet functions proactively. It connects to software tools, reviews notification feeds, sends daily morning summaries, and alerts users when items require direct intervention. Built on Anthropic's Opus 5.5 model, the system operates within a cloud-based virtual machine equipped with its own web browser to execute multi-step workflows.
Anthropic said the agent maintains a single continuous conversation thread rather than separate chat sessions, drawing on memory, custom skills, and connected tools established in a user's primary Claude account.
Workplace automation and safety controls
Claude Duet is tailored for enterprise administrative tasks, financial forecasting, dashboard creation, and custom web application development. Anthropic noted that because the tool focuses on workplace tasks, consumer-oriented capabilities such as booking personal tickets remain less developed.
To maintain security, the company stated that Claude Duet operates with explicit operational boundaries:
- The agent does not view user passwords or login credentials when accessing linked accounts.
- A dedicated classifier and safety prompt screen outgoing messages to prevent sensitive data leaks.
- Background tasks require explicit user confirmation before executing critical actions, including sending messages, making payments, signing documents, or deleting files.
Anthropic acknowledged that during the early preview phase, the model "might misread a thread or miscalibrate what's urgent."
Comparing proactive AI agents
| Agent | Developer | Target audience | Core features |
|---|---|---|---|
| Claude Duet | Anthropic | Professionals, small businesses | Morning briefs, Opus 5.5, cloud VM browser |
| Muse | Meta | Consumers | Personal agent, consumer app integration |
| Dots | OpenAI | Enterprise and consumers | Always-on background execution |
The unveiling follows rapid moves by rivals into proactive computing. Meta introduced its Muse agent last month, which Meta chief executive Mark Zuckerberg told analysts on Wednesday had been set up by more than 10 million people in the United States and Canada. OpenAI launched its own always-on agent, named Dots, weeks later during its annual developer conference.
Availability and subscription requirements
The preview of Claude Duet is accessible via a waitlist for Claude Pro and Max subscribers based in the United States. Users can access the service through the main Claude website or a dedicated standalone Claude Duet iOS application. Usage of the agent's background checks and automated tasks counts directly against a user's existing plan limits.
Key numbers
| What | Figure | Source |
|---|---|---|
| Target audience | Professionals and small businesses | Anthropic |
| Underlying model | Opus 5.5 | Anthropic |
| Meta Muse adoption | Over 10 million users in US and Canada | Mark Zuckerberg |
In this story
| Name | Role in the story |
|---|---|
| Anthropic | Developer that launched the Claude Duet AI agent preview. |
| Claude Duet | The proactive AI agent preview being introduced. |



